Why your follower keeps losing you: body blocking
The demo goes fine. You walk, it follows. You stop, it stops. Everyone nods.
Then someone turns sideways to grab a tool, steps behind a machine, or just shifts their weight so their body is between the tag and the vehicle. The whole thing falls apart. The cart hesitates, veers, or slows to a crawl like it forgot what it was doing.
We have hit this problem more times than we care to remember. It is not a software bug in the usual sense. It is physics. The industry calls it body blocking, or more broadly, non-line-of-sight interference. If you are evaluating any auto-follow system, this is the thing to test, because it will happen in every real deployment.
What is actually going wrong
Most follow systems use UWB for good reason. It gives you centimeter-level ranging, works indoors, and does not care about GPS dead zones. But it is still radio. Radio needs a path between the tag and the antennas on the vehicle, and the human body is surprisingly bad news at UWB frequencies.
The water in human tissue absorbs and attenuates the signal. When you turn and your torso gets between the tag and the vehicle, the direct path can weaken or vanish.
The signal does not politely stop, though. It finds other ways. It bounces off metal racks, walls, machines, or the forklift parked to your left. Those reflected paths arrive late and from the wrong direction. Now the system is working with a distance that looks reasonable but is biased, or a direction estimate that is quietly wrong.
A completely lost signal is actually the easy case. You detect it, you handle it. The dangerous case is the believable-but-wrong measurement. The system gets a number, trusts it, and acts on it. It does not know you just turned your shoulders. It only knows the coordinates jumped.
If the system uses angle-of-arrival for direction, this gets worse. NLOS can make the angle snap before the distance drifts much at all. That is the follower that suddenly yanks sideways instead of just slowing down.
In our tests, channel 9 is hit hardest: when the tag is close to the body and the person faces away, the direct path can be almost completely cut. Channel 5 behaves better, but it is still affected. Body blocking is not an edge case. It is one of the hardest problems in UWB following.
Why indoors makes it worse
Outdoors, body blocking is usually brief. You turn back, the signal returns, and following picks up again.
Indoors, you are surrounded by reflectors: metal equipment, conveyor frames, pipe racks, machinery housings, shutters, pallets, hospital beds, carts, doors, and other people. The direct path gets blocked, and several reflected paths are suddenly available. The receiver picks one up, and it is lying.
It is not just your own body, either. A colleague walking between you and the vehicle, a forklift crossing the path, or a door swinging open at the wrong moment can create the same problem.
Now you have two problems stacked: the direct signal is gone, and the fallback signal is misleading. If the system trusts whatever measurement came in last, it will make bad decisions at exactly the wrong moment, which is when the person is turning, stopping, or stepping around something.
It is not just a positioning problem
The instinct is to say, “Our UWB lost the signal, we need better UWB.” Sometimes that is the right call. Antenna placement matters. Tag orientation matters. Mounting height matters.
But in a real vehicle, body blocking is not just a positioning problem. It is a continuity problem. The question is not only, “Where is the person?” It is, “What does the vehicle do for the half-second when the direct signal is garbage?”
A clean coordinate in perfect conditions is not the hard part. The hard part is holding together when the direct path disappears.
A lot of products fall apart right here. The positioning module outputs coordinates. The controller waits for good coordinates. The coordinates go bad. The controller has nothing else to work with, so the vehicle stops, jerks, or follows a position that was never real.
How followwagon handles it
We built followwagon’s system specifically with this failure mode in mind.
The UWB layer, the following logic, the vehicle’s motion state, and the motor control all run in one real-time system. They share a time base. UWB ranging, inertial measurements, and encoder feedback are all stamped and compared as data from the same moment, not stitched together afterward.
We do not treat a raw measurement as truth. If a reading would require an impossible acceleration, or a sudden direction change that makes no physical sense given the person’s trajectory and the vehicle’s motion, it gets less weight. Kinematic constraints filter out measurements that cannot be real.
When the direct signal drops briefly, the system does not throw away everything it knows. It uses the person’s recent speed and heading, plus the vehicle’s own motion data, to carry a prediction through the gap. Not forever. Just through the normal moments when a shoulder, doorway, machine, or passer-by breaks the link.
When the signal comes back, the system blends it in. It does not snap hard toward the first fresh reading, which is sometimes still a bounce.
If the blockage does not resolve, the prediction stops and the safety layer takes over. The follower decelerates in a controlled way and stops, waiting to reacquire the person. Onboard obstacle sensing stays active independently of the UWB link. Lost-link is treated as a safety state, not an edge case.
What helps in practice
Some of this is on the product, but deployment and wearing habits matter more than people expect.
- Wear the tag where the vehicle can see it.A tag in a deep pocket, under a heavy coat, or behind a metal tool case will underperform. On the belt, outside a hip bag, on a shoulder strap, or on the outward face of a backpack usually works better than a chest-mounted position. Test it in the posture people actually use.
- Keep the tag away from metal.A phone leaned against it, a metal case, or a dense equipment load nearby all make the radio environment worse.
- Check the antennas on the vehicle.If the chassis, battery box, cables, or cargo are blocking them, small mounting changes can make a noticeable difference.
- Test turns, not just straight lines.Straight-line following is easy. Turn your shoulders. Stop hard. Walk around a corner. That is where weak systems show themselves.
- Test where it will actually be used.A parking lot tells you almost nothing. A factory with metal everywhere, a warehouse aisle, a hospital corridor, or a course with people walking through tells you everything.
- Watch the recovery, not only the dropout.A follower can stop safely during a dropout and still be unpleasant to use if it recovers with a lurch. A good follower comes back to stable tracking without lunging, over-steering, or going quiet for three seconds.
The test
Walk forward. Turn sideways. Stop. Walk again. Do it near metal equipment. Do it around a doorway. Have someone walk between you and the vehicle.
If the system pretends the radio link will always be clean, it was not built for real use. The link will break. The question is whether the vehicle knows what to do when it does.
Body blocking is unavoidable. Losing the person because of it is not.